Semi-Automated Construction of Adversarial Agents for Trainable Automated Forces.

Abstract

Virtual agents are used by the military for extensive training of pilots. However, creating virtual agents with the appropriate behaviors is a lengthy and specialized process. We describe a new tool (TAF) to allow rapid creation of virtual adversarial agents by non-specialized personnel. In TAF, users can draw diagrams of agent behavior which are matched within a pre-existing knowledge base. By combining several instances of similar behaviors a behavior model for a new agent is created that matches the diagram provided by the user.